South Pole tourism trend prediction system and method based on ship position big data

Abstract

The invention relates to the technical field of travel trend prediction, and discloses a South Pole travel trend prediction system and method based on ship position big data, and the system comprises a ship position data collection module, an environment data collection module and the like. Ship navigation data are collected through a satellite AIS system, multi-source data are obtained in combination with a meteorological satellite, an ocean buoy, a ship registration system and the like, the data are fused through a dynamic space-time grid coding technology, and space-time features are extracted to construct a dynamic feature matrix. And capturing a space-time coupling relationship by adopting a space-time diagram neural network, optimizing a parameter weight according to a historical error, and analyzing an abnormal event to generate an influence factor. And a prediction result output module superposes trend prediction and event influence factors and outputs a travel hotspot area probability distribution diagram. And the visual interaction module renders the result into a dynamic thermodynamic diagram and superposes the dynamic thermodynamic diagram to an electronic chart, and supports multi-layer display and time axis interaction. The Antarctic tourism trend can be accurately predicted, and powerful support is provided for tourism decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of tourism trend prediction, and particularly to an Antarctic tourism trend prediction system and method based on ship position big data. Background Art

[0002] With the booming development of global tourism, polar tourism has gradually emerged. Antarctica attracts more and more tourists with its unique natural scenery and ecological environment. However, due to the special geographical environment and climatic conditions in the Antarctic region, tourism activities face many challenges. Accurate tourism trend prediction is crucial for ensuring tourist safety, optimizing the allocation of tourism resources, and promoting the sustainable development of Antarctic tourism.

[0003] In the past, traditional tourism trend prediction methods mainly relied on statistical analysis and empirical judgment. Statistical analysis often depends on limited sample data and is difficult to comprehensively reflect the complex and changeable actual situation of Antarctic tourism. Empirical judgment, on the other hand, is greatly affected by subjective factors and lacks scientificity and accuracy. For example, when predicting the number of Antarctic tourists in the early stage, only the number of tourists in the same period of previous years was simply used to speculate on future trends, without considering various factors such as climate change, policy adjustments, and the development of new tourism projects that affect tourism trends, resulting in a large deviation between the prediction results and the actual situation.

[0004] With the development of information technology, some studies have begun to attempt to use data-driven methods for tourism trend prediction. However, in the field of Antarctic tourism, the difficulty of data acquisition is a key issue. The Antarctic region is vast and has a harsh environment, making it difficult to build infrastructure, which makes it difficult to widely deploy and stably operate data collection devices. At the same time, Antarctic tourism involves various types of data, such as ship navigation data, environmental data, tourist behavior data, etc. These data sources are scattered and have different formats. How to effectively integrate these multi-source heterogeneous data has become a difficult problem. In addition, existing prediction models have deficiencies in processing spatio-temporal data. Antarctic tourism trends have obvious spatio-temporal characteristics. The navigation trajectories of ships, the activities of tourists, and environmental factors all change continuously over time and space. Traditional machine learning models, such as linear regression and decision trees, are difficult to capture the spatio-temporal dependence relationships in the data and cannot accurately predict Antarctic tourism trends. Although some complex deep learning models can theoretically process spatio-temporal data, in practical applications, due to the scarcity and complexity of Antarctic tourism data, the models are prone to overfitting or underfitting phenomena, resulting in low prediction accuracy.

[0005] In terms of visualization and interaction, existing Antarctic tourism-related systems also have deficiencies. Many systems simply display data, lacking intuitive and dynamic visualization effects, and are unable to enable users to quickly and accurately understand the changes in tourism trends. Moreover, the interactivity between users and the system is poor, making it difficult to meet the personalized needs of different users. For example, tourism practitioners hope to intuitively understand future tourism hotspots through the system in order to plan tourism routes and arrange service facilities in advance; tourists are more concerned about tourism risk information and hope to query tourism situations in specific regions and time periods according to their own needs. However, existing systems cannot well meet these needs. Summary of the Invention

[0006] The purpose of the present invention is to provide a system and method for predicting Antarctic tourism trends based on ship position big data to solve the problems raised in the above background technology.

[0007] To achieve the above purpose, the present invention provides the following technical solution: A system for predicting Antarctic tourism trends based on ship position big data, the system includes an environmental data collection module, a tourist behavior collection module, a multi-source data fusion module, a spatio-temporal feature extraction module, a trend prediction model construction module, a dynamic weight adjustment module, an abnormal event analysis module, a prediction result output module, and a visualization interaction module; The ship position data collection module obtains the navigation trajectory data of ships in the Antarctic region in real time through the satellite AIS system, including ship position, speed, and heading information; The environmental data collection module integrates meteorological satellite and ocean buoy data to collect wind speed, sea ice coverage rate, and seawater temperature parameters in the Antarctic region; The tourist behavior collection module obtains tourist quantity, stay duration, and activity area distribution data through the ship registration system and mobile terminals; The multi-source data fusion module uses dynamic spatio-temporal grid coding technology to perform time-space alignment on ship position data, environmental data, and tourist behavior data to generate a multi-dimensional spatio-temporal fusion data set; The spatio-temporal feature extraction module extracts the spatial aggregation degree, heading consistency index, and environmental parameter change gradient of the ship trajectory based on the sliding time window mechanism to construct a dynamic feature matrix; The trend prediction model construction module uses a spatio-temporal graph neural network to capture the spatio-temporal coupling relationship between ship activities and environmental factors through gated spatio-temporal convolutional units; The dynamic weight adjustment module optimizes the weight allocation of environmental parameters and tourist behavior parameters using an adaptive particle swarm algorithm according to the root mean square value of historical prediction errors; The abnormal event analysis module identifies meteorological mutation events and abnormal ship aggregation phenomena by comparing real-time data with historical baselines, and generates event impact factors; The prediction result output module performs weighted superposition of the trend prediction result and the event impact factor, and outputs the probability distribution map of Antarctic tourism hot spots in the future period; The visualization and interaction module uses the WebGL engine to render the prediction result as a dynamic heat map and superimposes it on the Antarctic electronic chart.

[0008] Preferably, the multi-source data fusion module uses the dynamic spatio-temporal grid encoding technology to define a three-dimensional grid cell of latitude-longitude-time, with a side length of 0.1 degree × 0.1 degree × 1 hour; performs spatial interpolation on the ship trajectory data, and calculates the ship density index of each grid cell; performs temporal resampling on the environmental data, and calculates the wind speed gradient, sea ice coverage rate, and average seawater temperature within the grid cell; performs spatial aggregation on the tourist behavior data to generate the tourist density index of the grid cell.

[0009] Preferably, in the spatio-temporal feature extraction module, the spatial aggregation degree calculates the distribution probability of ships within the grid cell through the kernel density estimation algorithm; the heading consistency index is generated through the cosine similarity matrix of the ship heading angles; the environmental parameter change gradient uses the Sobel operator to calculate the difference in sea ice coverage rate between adjacent grid cells.

[0010] Preferably, the input layer of the spatio-temporal graph neural network receives the dynamic feature matrix, and the hidden layer contains gated spatio-temporal convolutional units, and its calculation formula is: ; where, is the hidden state matrix of the th layer, is the hidden state matrix of the th layer, is the spatio-temporal convolutional kernel, is the correlation matrix of ship activities and environmental parameters, is the weight matrix, represents the Hadamard product, is the activation function.

[0011] Preferably, in the abnormal event analysis module, the detection of meteorological mutation events is realized by calculating the Z-score of the real-time wind speed and the historical mean, and when is triggered, an event mark is generated; the abnormal aggregation phenomenon of ships is identified through the DBSCAN clustering algorithm. If the clustering radius is less than 5 nautical miles and the density exceeds the threshold, an event impact factor is generated.

[0012] Preferably, the calculation formula of the event impact factor is: ; where, is the difference between the real-time ship density and the baseline, is the average value of ship density in the same historical period, is the change amount of sea ice coverage rate, is the average value of sea ice coverage rate in the same historical period, 、 is the dynamic adjustment coefficient.

[0013] Preferably, the prediction result output module uses the Monte Carlo simulation method to superimpose the trend prediction result with the event impact factor to generate the probability distribution of the tourism risk level of each grid cell within the next 72 hours, and the probability value is normalized through the softmax function.

[0014] Preferably, the visualization and interaction module supports multi-layer superposition display, including the real-time ship trajectory layer, the sea ice distribution layer, and the predicted heat map layer; users can view the comparative analysis of historical predictions and real-time data by sliding the timeline.

[0015] Preferably, the present invention further includes a method for predicting the Antarctic tourism trend based on ship position big data, including the following steps: S1: Real-time collect the latitude, longitude, speed and heading data of ships in the Antarctic region through the satellite AIS system, and synchronously obtain the wind speed, sea ice remote sensing data of the meteorological satellite and the seawater temperature data transmitted by the buoy; S2: Adopt the dynamic spatio-temporal grid coding technology to interpolate the ship trajectory data into a 0.1-degree × 0.1-degree grid, and resample the environmental data to a 1-hour resolution in time series to generate a multi-source spatio-temporal fusion data set; S3: Extract the ship spatial aggregation degree, heading consistency index and sea ice coverage rate change gradient based on the sliding time window, and construct a dynamic feature matrix; S4: Train the spatio-temporal graph neural network model, input the dynamic feature matrix, and learn the spatio-temporal correlation between ship activities and environmental parameters through the gated spatio-temporal convolutional unit; S5: Use the adaptive particle swarm optimization algorithm to optimize the environmental parameter weights, and use the root mean square value of the historical prediction error as the fitness function to iteratively update the weight vector; S6: Real-time calculate the wind speed Z-score and ship clustering density, and generate an event impact factor when an abnormal event is detected; S7: Use the Monte Carlo simulation to superimpose the trend prediction result with the event impact factor, and output the tourism risk probability distribution of each grid cell within the future period; S8: Render the probability distribution as a dynamic heat map through the WebGL engine and integrate it with the electronic nautical chart to support users to interactively query the prediction details of a specific spatio-temporal range.

[0016] Preferably, the present invention further includes an electronic device for implementing the above-mentioned prediction system for the Antarctic tourism trend based on ship position big data, including: A processor configured to perform the following operations: receive vessel AIS data through a satellite communication module, and call a multi-source data fusion algorithm to perform spatio-temporal grid encoding on vessel position data, environmental data, and tourist behavior data; run a spatio-temporal graph neural network model to extract spatio-temporal coupling features of vessel activities and environmental parameters, and dynamically adjust prediction weights based on an adaptive particle swarm algorithm; analyze influence factors of abnormal events in real time and superimpose them on the prediction results to generate a probability distribution. A memory storing an executable instruction set, the instruction set including a dynamic spatio-temporal grid encoding program, a spatio-temporal feature extraction program, a weight optimization program, and an abnormal event detection algorithm. A communication module integrating a satellite communication unit and a 5G transmission unit, configured to obtain meteorological satellite data, ocean buoy data, and tourist behavior data of a vessel registration system, and transmit prediction results to a cloud server. A display module equipped with a WebGL rendering engine, configured to superimpose and display a prediction heat map and an electronic nautical chart, and support user interaction operations. A power supply module using a low-temperature environment adaptive power management chip to provide stable power supply for the processor, the memory, and the communication module. A hardware accelerator configured to perform parallel acceleration on convolution operations and Monte Carlo simulations of a spatio-temporal graph neural network through an FPGA chip to improve model inference efficiency.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention obtains multi-source data such as vessel navigation trajectories, environmental parameters, and tourist behaviors through a vessel position data acquisition module, an environmental data acquisition module, and a tourist behavior acquisition module. The dynamic spatio-temporal grid encoding technology is used for data fusion to align data from different sources and in different formats in time and space, generating a multi-dimensional spatio-temporal fusion data set. The comprehensive utilization of such multi-source data can more comprehensively reflect the actual situation of Antarctic tourism. Compared with traditional prediction methods that only rely on a single type of data, the prediction accuracy is greatly improved. For example, by combining vessel position and sea ice coverage data, it is possible to more accurately determine which areas are affected by sea ice changes and thus affect the distribution of tourist hotspots; considering data on tourist stay duration and activity area distribution can better predict changes in tourist demand and tourism trends.

[0018] Based on the sliding time window mechanism, the spatio-temporal feature extraction module extracts features such as the spatial aggregation degree, heading consistency index, and environmental parameter change gradient of ship trajectories, and constructs a dynamic feature matrix. The trend prediction model construction module uses a spatio-temporal graph neural network to capture the spatio-temporal coupling relationship between ship activities and environmental factors through gated spatio-temporal convolutional units. This advanced feature extraction and model construction method can effectively mine the complex spatio-temporal dependence relationships in the data and can predict the Antarctic tourism trend more accurately than traditional models. Taking the ship spatial aggregation degree and the environmental parameter change gradient as examples, they reflect the dynamic relationship between ship activities and environmental changes. The spatio-temporal graph neural network can learn these relationships and be used to predict the change trend of future tourism hotspots.

[0019] The dynamic weight adjustment module optimizes the weight allocation of environmental parameters and tourist behavior parameters using the adaptive particle swarm algorithm according to the root mean square value of historical prediction errors. As the data is continuously updated and the actual situation changes, the model can automatically adjust the parameter weights to make the prediction model more adaptable to the complex and changeable environment of Antarctic tourism, further improving the accuracy and stability of the prediction. For example, during the peak and off-peak seasons of Antarctic tourism, the influence degrees of environmental factors and tourist behavior factors on the tourism trend may be different. The dynamic weight adjustment mechanism can automatically adjust the weights according to historical data and prediction errors, enabling the model to maintain a high prediction accuracy in different seasons.

[0020] The abnormal event analysis module identifies meteorological mutation events and abnormal ship aggregation phenomena by comparing real-time data with historical baselines, and generates event impact factors. This function can timely detect abnormal situations that may affect Antarctic tourism and quantify their impact degrees. The prediction result output module performs weighted superposition on the trend prediction result and the event impact factor, and outputs a prediction result that is more in line with the actual situation, enhancing the reliability of the prediction. For example, when a meteorological mutation event is detected, the system can timely adjust the prediction result, issue a warning to tourism practitioners and tourists, and remind them to take corresponding countermeasures to ensure the safe conduct of tourism activities.

[0021] The visualization and interaction module uses the WebGL engine to render the prediction results as dynamic heat maps and superimpose them on the Antarctic electronic chart, supporting multi-layer superimposed display and timeline sliding operations. This intuitive and dynamic visualization method enables users to quickly and accurately understand the changes in Antarctic tourism trends. Tourism practitioners can plan tourist routes and arrange service facilities in advance based on the prediction results; tourists can more intuitively understand tourism risks and hot spots and reasonably plan their itineraries. For example, by viewing the dynamic heat map, tourism practitioners can intuitively see the distribution and changing trends of future tourism hot spots, allocate resources in advance, and improve service quality; by sliding the timeline to compare historical predictions and real-time data, tourists can better assess tourism risks and make more informed tourism decisions. At the same time, the multi-layer superimposed display and interaction functions meet the personalized needs of different users, enhancing the user experience and decision-making efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is the working principle diagram of the prediction system described in the present invention; Figure 2 is the data processing step diagram of the multi-source data fusion module; Figure 3 is the feature calculation diagram of the spatio-temporal feature extraction module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0024] Please refer to Figures 1 - 3 , the present invention provides a technical solution: a prediction system for Antarctic tourism trends based on ship position big data, which mainly consists of a ship position data acquisition module, an environmental data acquisition module, a tourist behavior acquisition module, a multi-source data fusion module, a spatio-temporal feature extraction module, a trend prediction model construction module, a dynamic weight adjustment module, an abnormal event analysis module, a prediction result output module, and a visualization and interaction module.

[0025] The ship position data acquisition module obtains the navigation trajectory data of ships in the Antarctic region in real time through the satellite AIS system, including ship position (latitude and longitude information), speed, and heading information. This module provides basic data for subsequent analysis of ship activities in the Antarctic region. For example, the ship's position information can reflect the specific Antarctic sea area it is in at different times, and the speed and heading are helpful for understanding the ship's movement trend.

[0026] The environmental data collection module integrates meteorological satellite and ocean buoy data to collect wind speed, sea ice coverage, and seawater temperature parameters in the Antarctic region. Meteorological satellites can obtain large-scale meteorological data from a macroscopic perspective, while ocean buoys can monitor relevant ocean parameters in real time in specific sea areas. These environmental data are crucial for analyzing the changes in the Antarctic tourism environment, as changes in wind speed, sea ice coverage, and seawater temperature directly affect the navigation safety of ships and the tourism experience of tourists.

[0027] The tourist behavior collection module obtains data on the number of tourists, stay duration, and activity area distribution through the ship registration system and mobile terminals. The ship registration system records the number of tourists carried on each voyage, while the mobile terminal can obtain the stay duration and activity area distribution of tourists in the Antarctic region with the help of positioning technology and relevant applications. These data can reflect the behavior characteristics of tourists during the Antarctic tourism process and provide an important basis for predicting tourism trends.

[0028] The multi-source data fusion module uses dynamic spatio-temporal grid coding technology to perform time-space alignment on ship position data, environmental data, and tourist behavior data, generating a multi-dimensional spatio-temporal fusion dataset. By defining specific three-dimensional grid cells, this module integrates data from different sources, enabling subsequent analysis to be carried out within a unified spatio-temporal framework.

[0029] The spatio-temporal feature extraction module extracts the spatial aggregation degree, course consistency index, and environmental parameter change gradient of ship trajectories based on the sliding time window mechanism, constructing a dynamic feature matrix. By extracting these features, the changing patterns of ship activities and environmental factors over time can be better captured, providing more representative input data for the trend prediction model.

[0030] The trend prediction model construction module uses a spatio-temporal graph neural network to capture the spatio-temporal coupling relationship between ship activities and environmental factors through gated spatio-temporal convolutional units. This model can make full use of the spatio-temporal information in the data to accurately predict the Antarctic tourism trend.

[0031] The dynamic weight adjustment module optimizes the weight allocation of environmental parameters and tourist behavior parameters using an adaptive particle swarm algorithm based on the root mean square value of historical prediction errors. By continuously adjusting the parameter weights, the accuracy of the prediction model can be improved, enabling it to better adapt to the complex and changing Antarctic tourism environment.

[0032] The abnormal event analysis module identifies meteorological mutation events and abnormal ship aggregations by comparing real-time data with historical baselines, generating event impact factors. This module can promptly detect abnormal situations that may affect Antarctic tourism and quantify their impact levels, providing more comprehensive information for the prediction results.

[0033] The prediction result output module performs weighted superposition on the trend prediction result and the event impact factor, and outputs the probability distribution map of the Antarctic tourism hotspots in the future period. This module combines the result of the prediction model with the impact of abnormal events, providing more valuable prediction information for tourism practitioners and tourists.

[0034] The visualization and interaction module uses the WebGL engine to render the prediction result as a dynamic heat map and superimposes it on the Antarctic electronic chart. This module provides an intuitive and convenient interaction interface for users, facilitating them to view the prediction result and conduct relevant analysis.

[0035] The present invention will be further described below in conjunction with Embodiments 1 to 5:

[0036] Embodiment 1:

[0037] In this embodiment, the multi-source data fusion module adopts the dynamic spatio-temporal grid coding technology, defining a three-dimensional grid cell of latitude-longitude-time with a side length of 0.1 degree × 0.1 degree × 1 hour.

[0038] For the ship trajectory data, due to the discreteness of the collection time and location, spatial interpolation is required. The inverse distance weighted interpolation method (IDW) is used for spatial interpolation to calculate the ship density index of each grid cell. Assuming that at a certain moment, several surrounding ship position points are known ( ) and their corresponding ship numbers , for the target grid cell , its ship density index is calculated by the formula: ;

[0039] where represents the distance between the target grid cell and the known ship position points. In this way, the ship trajectory data can be evenly distributed into each grid cell, facilitating subsequent analysis of the ship distribution in different regions.

[0040] For environmental data, such as wind speed, sea ice coverage rate, and seawater temperature, etc., due to the inconsistent collection frequency and time interval, time series resampling is required. Taking the wind speed data as an example, it is resampled to a 1-hour resolution using the linear interpolation method. Assuming that the wind speeds collected at time points and are and respectively, for the time point ( ), the resampled wind speed is calculated by the formula: ;

[0041] Then calculate the wind speed gradient within the grid cell. The wind speed gradient reflects the spatial variation of the wind speed. By calculating the ratio of the wind speed difference between adjacent grid cells to the grid spacing, the wind speed gradient is obtained. At the same time, calculate the sea ice coverage rate and the average seawater temperature within the grid cell. For the sea ice coverage rate, directly count the proportion of the area covered by sea ice within the grid cell; for the seawater temperature, perform an arithmetic average of the seawater temperatures at all sampling points within the grid cell.

[0042] For the tourist behavior data, perform spatial aggregation. According to the information of tourists in the ship registration system and the location information obtained from the mobile terminal, assign tourists to the corresponding grid cells. By counting the number of tourists in each grid cell and combining it with the area of the grid cell, generate the tourist density index of the grid cell. Suppose the area of a certain grid cell is and the number of tourists in this grid cell is , then the calculation formula for the tourist density index is: ;

[0043] Through the above processing of ship trajectory data, environmental data, and tourist behavior data, a multi-dimensional spatio-temporal fusion data set is generated, providing rich and accurate data support for subsequent analysis and prediction.

[0044] Example 2:

[0045] This example mainly describes the specific calculation methods of spatial aggregation degree, course consistency index, and environmental parameter change gradient in the spatio-temporal feature extraction module. The accurate extraction of these features helps to more deeply analyze the change laws of ship activities and environmental factors, and improve the accuracy of trend prediction.

[0046] In the spatio-temporal feature extraction module, the spatial aggregation degree is calculated by the kernel density estimation algorithm for the distribution probability of ships within the grid cell. Taking the Gaussian kernel function as an example, assume that the position points of ships within the grid cell are ( ), for the target position , its spatial aggregation degree is calculated as: ;

[0047] Among them, is the bandwidth parameter, is the Gaussian kernel function. Through the kernel density estimation algorithm, it can more accurately reflect the aggregation degree of ships within the grid cell, rather than simply counting the number of ships.

[0048] The course consistency index is generated through the cosine similarity matrix of ship course angles. Suppose at a certain moment, there are A ship, with heading angles of . First, calculate the cosine similarity matrix of the ship's heading angles , where ( ). Then, perform an average calculation on the cosine similarity matrix to obtain the heading consistency index , and the calculation formula is: ;

[0049] The heading consistency index reflects the degree of consistency of the ship's headings within the grid cell. The higher the index, the more consistent the ship's headings are, indicating that there may be a certain common behavior pattern or influencing factor.

[0050] The environmental parameter change gradient uses the Sobel operator to calculate the difference in sea ice coverage between adjacent grid cells. Taking the two-dimensional Sobel operator as an example, assume that the matrix formed by the sea ice coverage data is , and use the Sobel operators and to perform convolution operations in the horizontal and vertical directions respectively.

[0051] ;

[0052] Calculate the gradient in the horizontal direction and the gradient in the vertical direction: ;

[0053] ;

[0054] Then calculate the sea ice coverage change gradient :

[0055] By calculating the sea ice coverage change gradient, the spatial change trend of the sea ice coverage can be captured, providing an important basis for analyzing the impact of environmental factors on ship activities and tourism trends.

[0056] Example 3:

[0057] This example details the calculation process of the gated spatio-temporal convolutional unit in the spatio-temporal graph neural network, which helps to understand how the trend prediction model learns the spatio-temporal correlation between ship activities and environmental factors, so as to achieve accurate prediction of Antarctic tourism trends.

[0058] The input layer of the spatio-temporal graph neural network receives the dynamic feature matrix constructed by the spatio-temporal feature extraction module. The hidden layer contains gated spatio-temporal convolutional units, and its calculation formula is: ;

[0059] Among them, is the -th layer hidden state matrix, is the -th layer hidden state matrix, is the spatio-temporal convolution kernel, which is used to extract spatio-temporal features; is the correlation matrix of ship activities and environmental parameters, which reflects the mutual relationship between ship activity characteristics and environmental parameters and is obtained through the analysis and learning of historical data; is the weight matrix, which controls the importance of different features in the model; represents the Hadamard product, which is used for element-wise multiplication; is the activation function. Here, the ReLU function is adopted, and its expression is .

[0060] During the model training process, by continuously adjusting the parameters of the spatio-temporal convolution kernel , the correlation matrix and the weight matrix , the model can better capture the spatio-temporal coupling relationship between ship activities and environmental factors. For example, at the beginning of training, these parameters are randomly initialized, and then according to the input dynamic feature matrix, the output of the hidden layer is calculated according to the above formula. By comparing the model prediction results with the actual data, the loss function (such as the mean square error loss function) is calculated. Using the backpropagation algorithm, the parameters are updated according to the gradient information of the loss function, so that the loss function gradually decreases and the prediction accuracy of the model continuously improves. After multiple iterations of training, the model can learn the complex spatio-temporal patterns in the data, so as to effectively predict the Antarctic tourism trend.

[0061] Example 4:

[0062] In the abnormal event analysis module, the detection of meteorological abrupt events is realized by calculating the Z-score of the real-time wind speed and the historical mean. Suppose the historical wind speed data is , its mean is , the standard deviation is , and the real-time wind speed is , then the calculation formula of the Z-score is: ;

[0063] When , an event flag is triggered, indicating that a significant mutation has occurred in the current wind speed, and this mutation may have a greater impact on ship navigation and tourist activities.

[0064] The abnormal aggregation phenomenon of ships is identified by the DBSCAN clustering algorithm. The DBSCAN algorithm is a density-based spatial clustering algorithm that classifies data points into core points, boundary points, and noise points. In this embodiment, taking the ship positions as data points, the clustering radius is set to 5 nautical miles, and the density threshold is obtained based on the statistics of historical data. If the clustering radius formed by the ships in a certain area in the clustering result of the DBSCAN algorithm is less than 5 nautical miles and the density exceeds the threshold, it is considered that there is an abnormal aggregation phenomenon of ships in this area.

[0065] When a meteorological mutation event or an abnormal aggregation phenomenon of ships is detected, it is necessary to generate an event impact factor. The calculation formula of the event impact factor is: ;

[0066] where, is the difference between the real-time ship density and the baseline, is the average value of the ship density in the same historical period, which is obtained by statistically analyzing the ship density in the same time period of historical data; is the change amount of the sea ice coverage rate, that is, the difference between the real-time sea ice coverage rate and the sea ice coverage rate at the previous moment; is the average value of the sea ice coverage rate in the same historical period; , are dynamic adjustment coefficients, and their values are determined through experiments or experience according to different environments and tourism scenarios to adjust the relative importance of the ship density change and the sea ice coverage rate change to the event impact factor. By calculating the event impact factor, the influence degree of abnormal events on the Antarctic tourism trend can be quantified, providing a basis for adjusting the subsequent prediction results.

[0067] Embodiment 5:

[0068] This embodiment mainly describes how the prediction result output module generates the probability distribution of the tourism risk level of each grid cell in the future period, and how the visual interaction module presents the prediction result to the user in an intuitive way and supports the user to perform interactive operations, facilitating the user to obtain and analyze the prediction information.

[0069] The prediction result output module uses the Monte Carlo simulation method to superimpose the result obtained by the trend prediction model and the event impact factor generated by the abnormal event analysis module to generate the probability distribution of the tourism risk level of each grid cell within the next 72 hours. The Monte Carlo simulation is a statistical method based on random sampling, which approximates the true probability distribution through a large number of repeated random trials.

[0070] Specifically, when operating, the trend prediction model will output the predicted value of the basic tourism popularity of each grid cell at different future times, denoted as . The event impact factor obtained by the abnormal event analysis module is , it will have an adjustment effect on the basic tourism popularity. For each grid cell, during the Monte Carlo simulation process, the number of simulations is set to (for example times). Each time a simulation is performed, based on the and of the current grid cell, combined with a certain random fluctuation (simulating the uncertainty in the actual situation), a temporary tourism popularity value is calculated. The random fluctuation here can be generated by a random function such as a normal distribution. Assuming the random fluctuation value is , its calculation formula is: , where follows a normal distribution with a mean of 0 and a standard deviation set according to the historical data fluctuation situation.

[0071] After calculating , according to the pre-set tourism risk level classification standard, is mapped to the corresponding risk level. For example, it is set that a tourism popularity value less than is a low-risk level, between and is a medium-risk level, and greater than is a high-risk level. After completing simulations, count the number of times each risk level appears , , (corresponding to low, medium, and high risk levels respectively), and then calculate the probabilities of each risk level , , , and the calculation formula is: , , .

[0072] Since the calculated probability values may not meet the normalization requirements of the probability distribution (that is, the sum of probabilities is not 1), it is necessary to perform normalization processing through the softmax function. The expression of the softmax function is: , where is the unnormalized probability value (that is, the , , calculated above), is the total number of risk levels (here ), is the normalized probability value. After being processed by the softmax function, the sum of the probabilities of each risk level is 1, which can accurately reflect the probability distribution of each grid cell being in different tourism risk levels within the next 72 hours and provide a scientific basis for tourism-related decisions.

[0073] The visualization interaction module uses the WebGL engine to render the prediction results into a dynamic heat map and superimpose it on the Antarctic electronic chart. WebGL is a JavaScript-based graphics rendering technology that can efficiently draw complex graphics and animations in web browsers.

[0074] During the visualization process, according to the probability distribution of the tourism risk levels of each grid cell obtained from the prediction result output module, different risk level probabilities are mapped to different colors. For example, the color corresponding to the low risk level is set to green, and its color value can be expressed as RGB(0, 255, 0); the medium risk level corresponds to yellow, and the color value is RGB(255, 255, 0); the high risk level corresponds to red, and the color value is RGB(255, 0, 0). At the same time, the transparency of the color is set according to the probability size. The higher the probability, the darker the color (the lower the transparency), so that the high and low levels of tourism risk in different regions can be more intuitively displayed.

[0075] These grid cells with color information are drawn on the Antarctic electronic chart to form a dynamic heat map. To achieve the dynamic effect, the visualization interaction module will continuously update the heat map according to the time series. For example, at hourly intervals, the changes in the tourism risk distribution at different times within the next 72 hours are shown, and users can clearly see the dynamic changes in the tourism risk areas over time.

[0076] The visualization interaction module supports multi-layer superimposed display, including the real-time ship track layer, sea ice distribution layer, and predicted heat map layer. The real-time ship track layer shows the real-time navigation track of ships in the form of lines and markers on the electronic chart by obtaining the ship position information transmitted in real time by the ship position data acquisition module, facilitating users to understand the actual operation of ships in the current Antarctic region. The sea ice distribution layer represents the distribution range and density of sea ice on the electronic chart with different colors or patterns according to the sea ice coverage data obtained by the environmental data acquisition module, providing users with information on the current sea ice conditions in the Antarctic sea area. The predicted heat map layer is the dynamic heat map generated according to the probability distribution of the tourism risk levels mentioned above. It is superimposed with the real-time ship track layer and the sea ice distribution layer, and users can comprehensively analyze the relationship between ship activities, sea ice conditions, and tourism risks.

[0077] Users can slide the timeline to view the comparative analysis of historical predictions and real-time data. In the visualization interface, a timeline control is set, and the range of the timeline covers the time period of historical data records and the time period of future predictions. When the user slides the timeline, the visual interaction module will obtain the historical prediction data and real-time data at the corresponding moment from the database according to the time point selected by the user and display them on the electronic chart. For example, at a certain moment, the user can simultaneously view the heat map of tourism risk predicted historically at that moment, the actual real-time track of ships, and the sea ice distribution. Through comparative analysis, the accuracy of the prediction model can be evaluated, and the differences and connections between the current tourism trend and the historical situation can be better understood, so as to make more reasonable tourism plans and decisions.

[0078] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0079] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An Antarctic tourism trend prediction system based on ship position big data, characterized in that: include: The ship position data acquisition module obtains the navigation track data of ships in the Antarctic region in real time through the satellite AIS system, including ship position, speed and heading information; Environmental data acquisition module, which integrates meteorological satellite and ocean buoy data to collect wind speed, sea ice coverage and seawater temperature parameters in the Antarctic region; The tourist behavior collection module obtains the number of tourists, length of stay and activity area distribution data through the ship registration system and mobile terminals; The multi-source data fusion module uses dynamic spatiotemporal grid coding technology to perform time-space alignment on ship position data, environmental data and tourist behavior data to generate a multi-dimensional spatiotemporal fusion data set; The spatiotemporal feature extraction module extracts the spatial aggregation degree, heading consistency index and environmental parameter change gradient of ship trajectories based on the sliding time window mechanism, and constructs a dynamic feature matrix; The trend prediction model building module adopts a spatiotemporal graph neural network to capture the spatiotemporal coupling relationship between ship activities and environmental factors through gated spatiotemporal convolution units; The abnormal event analysis module compares real-time data with historical baselines to identify sudden meteorological events and abnormal ship aggregation, and generates event impact factors; The forecast result output module performs weighted superposition of trend forecast results and event impact factors, and outputs a probability distribution map of Antarctic tourism hotspots in the future period.

2. The Antarctic tourism trend prediction system according to claim 1, characterized in that: Also includes: The dynamic weight adjustment module uses the adaptive particle swarm algorithm to optimize the weight distribution of environmental parameters and tourist behavior parameters according to the root mean square value of historical prediction errors; The visualization interaction module uses the WebGL engine to render the prediction results into dynamic heat maps and overlay them on the Antarctic electronic chart.

3. The Antarctic tourism trend prediction system according to claim 1, characterized in that: The multi-source data fusion module adopts dynamic spatiotemporal grid coding technology to define latitude-longitude-time three-dimensional grid units with a side length of 0.1 degree × 0.1 degree × 1 hour; spatially interpolates ship trajectory data to calculate the ship density index of each grid unit; performs time series resampling on environmental data to calculate the wind speed gradient, sea ice coverage and seawater temperature mean within the grid unit; and spatially aggregates tourist behavior data to generate a tourist density index for the grid unit.

4. The Antarctic tourism trend prediction system according to claim 3, characterized in that: In the spatiotemporal feature extraction module, the spatial aggregation degree is calculated by the kernel density estimation algorithm to calculate the distribution probability of ships in the grid cells; the heading consistency index is generated by the cosine similarity matrix of the ship heading angles; and the environmental parameter change gradient is calculated by the Sobel operator to calculate the difference in sea ice coverage between adjacent grid cells.

5. The Antarctic tourism trend prediction system according to claim 4, characterized in that: The input layer of the spatiotemporal graph neural network receives the dynamic feature matrix, and the hidden layer contains a gated spatiotemporal convolutional unit, which is calculated as follows: ； in, For the layer hidden state matrix, For the layer hidden state matrix, is the spatiotemporal convolution kernel, is the correlation matrix between ship activities and environmental parameters, is the weight matrix, represents the Hadamard product, is the activation function.

6. The Antarctic tourism trend prediction system according to claim 5, characterized in that: In the abnormal event analysis module, the detection of meteorological mutation events is achieved by calculating the Z-score of the real-time wind speed and the historical mean. When the event marker is triggered, To calculate the Z-score value obtained; the abnormal aggregation of ships is identified by the DBSCAN clustering algorithm. If the cluster radius is less than 5 nautical miles and the density exceeds the threshold, the event impact factor is generated.

7. The Antarctic tourism trend prediction system according to claim 6, characterized in that: The calculation formula of the event impact factor is: ； in, is the difference between the real-time ship density and the baseline, is the average ship density during the same period in history, is the change in sea ice coverage, is the average sea ice coverage rate during the same period in history. , is the dynamic adjustment coefficient.

8. The Antarctic tourism trend prediction system according to claim 1, characterized in that: The prediction result output module adopts the Monte Carlo simulation method to superimpose the trend prediction results with the event impact factors to generate the probability distribution of the tourism risk level of each grid unit in the next 72 hours, and the probability value is normalized by the softmax function.

9. A method for predicting Antarctic tourism trends based on ship position big data, applied to the Antarctic tourism trend prediction system based on ship position big data as claimed in any one of claims 1 to 8, characterized in that: The following steps are involved: S1: The satellite AIS system collects the latitude, longitude, speed and heading data of ships in the Antarctic region in real time, and simultaneously obtains the wind speed and sea ice remote sensing data from meteorological satellites and the sea water temperature data transmitted by buoys; S2: Using dynamic spatiotemporal grid coding technology, the ship trajectory data is interpolated to a 0.1 degree × 0.1 degree grid, and the environmental data is time-series resampled to a 1-hour resolution to generate a multi-source spatiotemporal fusion dataset; S3: Based on the sliding time window, the spatial aggregation of ships, the heading consistency index and the gradient of sea ice coverage are extracted to construct a dynamic feature matrix; S4: Train the spatiotemporal graph neural network model, input the dynamic feature matrix, and learn the spatiotemporal correlation between ship activities and environmental parameters through gated spatiotemporal convolution units; S5: Use the adaptive particle swarm algorithm to optimize the weights of environmental parameters, and iteratively update the weight vector using the root mean square value of the historical prediction error as the fitness function; S6: Calculate wind speed Z-score and ship clustering density in real time, and generate event impact factors when abnormal events are detected; S7: Monte Carlo simulation is used to superimpose trend prediction results with event impact factors to output the tourism risk probability distribution of each grid cell in the future period.

10. An electronic device for implementing the Antarctic tourism trend prediction system based on ship position big data as claimed in any one of claims 1 to 8, characterized in that: include: The processor is configured to perform the following operations: receiving ship AIS data through a satellite communication module, calling a multi-source data fusion algorithm to perform spatiotemporal grid encoding on ship position data, environmental data, and tourist behavior data; running a spatiotemporal graph neural network model to extract spatiotemporal coupling characteristics of ship activities and environmental parameters, and dynamically adjusting prediction weights based on an adaptive particle swarm algorithm; Analyze the influencing factors of abnormal events in real time and superimpose them with the prediction results to generate probability distribution; A memory storing an executable instruction set, wherein the instruction set includes a dynamic spatiotemporal grid encoding program, a spatiotemporal feature extraction program, a weight optimization program, and an abnormal event detection algorithm; The communication module integrates a satellite communication unit and a 5G transmission unit to obtain meteorological satellite data, ocean buoy data, and visitor behavior data from the ship registration system, and transmit the prediction results to the cloud server; The display module is equipped with a WebGL rendering engine, which is used to overlay the predicted thermal map with the electronic chart and support user interactive operations; A power module, using a low temperature environment adaptive power management chip, to provide a stable power supply for the processor, memory and communication module; The hardware accelerator is configured to perform parallel acceleration of the convolution operations and Monte Carlo simulations of the spatiotemporal graph neural network through FPGA chips to improve the efficiency of model reasoning.

Images